Table 2.

Performance of prediction models for predicting the PNALD in patients with CIF in the training set.

Model	Sensitivity	Specificity	Accuracy	Youden index	AUC (95% CI)	p-valuea
Combined model	0.895	0.740	0.781	0.635	0.889 (0.836, 0.942)	NA
Radiomics model	0.632	0.827	0.776	0.459	0.776 (0.699, 0.852)	<0.001
Clinical model	0.754	0.728	0.735	0.483	0.826 (0.767, 0.885)	0.010
Unet	0.894	0.815	0.836	0.710	0.927 (0.894, 0.960)	0.102
ResNet+XGboost/FC	0.859	0.802	0.817	0.662	0.874 (0.875, 0.954)	0.327
VIT	0.789	0.741	0.753	0.530	0.834 (0.775, 0.892)	<0.001
SwinUNETR	0.824	0.728	0.753	0.553	0.832 (0.773, 0.892)	<0.001
DenseNet121	0.807	0.784	0.790	0.591	0.856 (0.801, 0.912)	0.151
ResNet18	0.772	0.778	0.776	0.550	0.840 (0.780, 0.899)	0.036
CNN	0.702	0.784	0.762	0.486	0.824 (0.764, 0.884)	0.008
MLP	0.737	0.846	0.817	0.582	0.864 (0.811, 0.916)	0.328

^aThe AUC of the combined model was compared with that of the other models using Delong test. Differences were considered statistically significant at p < 0.05.